3 replies. Last post: 2009-01-27Reply to this topic Return to forum
I’m looking for some new ideas for static evaluation function. A function which tells you what is the most likely output (result) of the game, without analyzing position in-depth. Please tell me if you have any ideas.
My current approach (not implemented yet, its tested on paper only) - minimal distance + kinda decomposition with articulation points (guards, sure territory etc)
You may want to take a look at Jens Lieberum’s paper on Amazong’s evaluation function.
Thank, i’ve already read it (4-5 times at least). I liked it. The main drawback is the lack of strategic knowledge. Combination of min-distance, queen mobility and king-distance is really good in practice. But often very good moves are missed, which are rather obvious for human.
Oh, one more thing, its very typical, that in some position you can chose simple moves that secures a win with small margin, and many others, more risky, that are evaluated higher. I’d rather chose simpler, less greedy, but sure move.